TCGdex MCP has well-structured tool definitions with clear naming (verb_noun pattern: search_cards, get_card, list_sets, get_set) and solid descriptions (avg 180 chars, within 10-1024 baseline). All tools have input schemas with type declarations. However, parameter descriptions are sparse, search_cards lacks descriptions for 'name' and 'limit' parameters; list_sets has empty properties. Output schemas are not documented, forcing LLMs to infer response structure. Error handling returns structured error objects but lacks recovery guidance. No tool annotations (readOnlyHint, idempotentHint) despite all tools being read-only. Response payloads are well-shaped (stripped API verbosity, include chaining IDs like card id/set id), but missing pagination guidance for potentially large result sets.

• Add descriptions to search_cards parameters: name → 'Card name or partial name to search for (e.g., "pikachu"). Omit to list all cards.' limit → 'Max cards to return (default 30, max 100).'
• Document output schemas for all tools. E.g., search_cards returns {count: number, cards: [{id, localId, name, image}]}. get_card returns {id, localId, name, category, rarity, hp, types, stage, set, illustrator, attacks: [{name, cost, damage, effect}], abilities, image_url}.
• Add tool annotations to inputSchema: readOnlyHint: true and idempotentHint: true for all four tools. This signals to agents that these tools are safe to retry.
• Enhance error messages with recovery guidance. E.g., if get_card fails with 404, return {error: 404, message: 'Card not found. Try search_cards(name="...") to find the correct id.'}
• Consider adding pagination support to search_cards and list_sets. If TCGdex API supports pagination, expose limit and offset/cursor parameters and document total_count in response.